AI Researcher at Alt Inc

Human + AI > AI

Here's the thing: making an AI that truly represents someone is hard. Not "fine-tune on their tweets" hard. More like "figure out what makes someone themselves and encode it" hard.

You'll be working on the technical challenges that make Alts possible. How do you extract decision-making patterns from messy artifacts like code reviews, voice notes, and the five different ways someone explains the same idea? What signals in someone's communication actually matter, and what's just noise?

The representation problem gets weird fast. Compress someone's personality and judgment into something computationally useful, but keep enough fidelity that their Alt still feels like them. Figure out where the line is between amplifying someone's voice and replacing it.

Then there's the interaction layer. An Alt needs to know when to respond on someone's behalf and when to stay quiet. When to mentor like its user would and when to defer. Context matters more than we've typically asked AI to handle.

What we're looking for:

You're the kind of person who reads psychology papers for fun and immediately thinks about architectures. You've probably built something with LLMs and been frustrated by how they can mimic style but miss the deeper patterns of how someone thinks.

You're comfortable with ambiguous metrics. "Does this feel like the person?" is a research question that requires talking to humans, not just optimizing loss functions. You may or may not have papers, but you definitely have opinions about why most benchmarks miss what matters.

You like working with personal, messy data. The kind where someone's judgment is scattered across Slack messages, decision docs, and casual conversations. You're good at finding signal in chaos.

You ship quickly. Research prototypes are fun, but you want to see your work in production with real users testing whether their Alts actually represent them.

You have taste about AI interactions. You've noticed why some AI feels natural and some feels uncanny, even when both are technically correct. You probably have strong opinions about anthropomorphization.

The work:

Architectures for capturing individual personality and judgment. Extraction systems for unstructured personal data. Building Alts that can represent someone authentically enough to mentor and connect on their behalf.

You'll spend time in papers, in prototypes, in conversations with users about whether their Alt "gets" them. We're early and figuring things out one experiment at a time.

If you're curious about making AI deeply personal and want to build with good people, let's talk.

We are funded, remote-friendly and have offices in Bangalore and Goa.